The purpose of this experiment was to apply to wisdom I gained from experiments preceeding my orals paper to see if I could improve the analytical precision of measurements on our instrument.
Here is a table detailing instrument settings. Reference it when needed.
No one is bragging about (or even reporting) their thorium ages, and adding another element to the chain in the sweep detracts from time we could use to measure more important ones.
Even in our experiments with unknowns, the 204Pb counts never exceeded the background by more than 3x, making them useless. Because we can't measure 204, measuring 202 is also useless. This is because we only measure 202Hg to account for a 204Hg (11% abundance) interference with 204Pb. Perhaps in future experiments with unknowns, it would be important to track 204Pb, even if we will never use it. But for these standard experiments where we want to prove we have very good measurements, I'm not including it.
In our previous experiments we dwelled on U238, 232Th, and 208Pb for 400 us, 206Pb for 800 us, and 207 for 2 ms. This yielded 206/238 ages that were noticably less precise than the 207/235 ages. For this experiment, we are assigning all elements the same dwell time.
Louise said that there isn't a significant reason to carefully partion dwell times unless the abundance of one of your elements is significantly different that others. The difference between U and Pb concentrations in our zircons is about a factor of 100. For example, if you are also measuring a major element alongside you something like U-Pb, you might want to turn the dwell time on that element way down to avoid overwhelming the detector.
A 2 ms dwell time is further desirable in that it is substantially higher than dwells I used for other experiments. For most analyses in my earlier work, 238U dwell was 1/5 that, and 206Pb 2/5 that. Longer dwells should improve the measurements. On those elements and increase precision.
The counts for U238 were being attenuated by a factor of ~250 by a little grated piece of metalk near the detector. Especially for some of the lower abundance samples, this reduction in counts really hampered our signal stability and precision. Removing the attenuator should improve this, so long as the counts don't go above 5 million. Spoiler: they didn't! And Plesovice, one of our standards, So that's good.
As best as I and Louise can tell, increasing the n of sweeps/cycle basically increases the number of measurements the machine will take before it reports an average. The minimum is 10. Before we were running samples with 20 or 30 sweeps/cycle. My logic is the same as Sidney's logic for the E361 XRF scans, why let the machine average your data for you when you could take the highest resolution data and look at it yourself? Collectively, the changes on the sweeps/cycle and dwell times have a new effect of shortening cycle duration (now < .1 seconds/cycle). In my opinion, analyzing the signal more frequently should be an advantage.
Below is a table of the names, ages, and sources for zircon substandards used in this study. It is the same set as the one used for my orals paper, but with the addition of the Mud Tank zircon, which comes from a pegmatitic carbonatite deposit in Australia.
For each run, of which there were ~10, the primary standard, FC-1 was analyzed 3 times at the beginning. Each zircon substandard was run in sequence. Each run finished with three more FC-1 analyses. This is perhaps overkill with the standardization, but on some of my runs from last year, I was at least wishing I had replicates. Below is a the evolution of counts throughout a run. 238U is in green and 206Pb is in pink. Counts for both axes in this axes are scaled the same. The run order is 3 FC-1, 91500, 9435, Oracle, OG-1, Tan-Br, Plesovice, Temora2, then 3 more FC1.
Below is the same graph, but with 206Pb scaled 1/10th of 238U, just to display more subtleties in Pb abundance.
For each sample, the laser was off for 23 seconds to allow flushing of sample chamber (seems like that takes 7-10 seconds) and to assess background levels. The laser is on for analysis for 7 seconds. I chose not to do a pre-ablation pass for reasons I'll elaborate on in the section below.
Because of the down-hole evolution of the U-Pb ratio, correctly approximating "beam seconds" is essential to the calculation of the age of a U-Pb laser sample. Iolite uses the term "beam seconds" as a calculated parameter for time since a "laser-on" event. This parameter is used by the software to align analyses in the depth/time domain.
Iolite creates a downhole fractionation model for your unknowns using time aligned analyses of your standards. You can see what this looks like below.
The software allows you to trim the noisy beginnings and ends of these data. You can also choose to fit the resulting data with a linear model, eponential model, spline, and more. You can also individually assess the downhole model for different isotope systems.
For this experiment, it looks like the Pb-U starts out high, perhaps because of Pb's higher volatility and lightness. It takes almost two seconds for it to the data to assume a linearly increasing Pb-U ratio. This seems like kind of a long time. I also trim off the end of the series as well, anything longer than the 7 seconds of "laser-on" time for the run. As for the fractionation vs. depth model, I went with the simplest option, a linear model. This fractionation-time model is used to calculate the Pb-U ratio at time zero, when the laser turns on. Mis-estimating time zero could cause systematic problems with accuracy.
Iolite also applies this same model, using beam seconds, to your samples with unknown ages. Therefore, it's important to have your beam seconds calibrated correctly.
Usually, calculating beam seconds is done with a laser log. This file, produced by the laser software, timestamps laser on/off signals, which you can synchronize to you mass spec data for flawless tracking of beam seconds. Hearing about this, I emailed some folks at Elemental Scientific to see if this was possible. However, according to engineers our hardware is too old to produce laser logs. They mentioned we could possibly achieve the same sort of binary signal by monitoring a major element in the mineral, but I can't think of any good element within 20% of our park mass (215).
Alternatively, Iolite allows you to approximate beam seconds by setting a threshold value to trigger a "laser-on" event and start counting beam seconds. I messed with this, and I decided when the counts for "Total Beam" increase above 2000 counts, we can start counting on Beam Seconds. I can also tell the software to not allow for more than 7 seconds of Beam Seconds On. This is why I chose not to do the pre ablation. The preablation sends another pulse of signal to the mass spec > than 2000, and it confuses Iolite's Beam Seconds algorithm. This is what a decent example of Beam Seconds (red) looks like for one of my runs.
The green is 238U counts, clipped to 10000 counts so you can see how well the algorithm times beam on. Something isn't right with duration. Total Beam Seconds hovers around 15. This is apparently how long it takes for the signal to go below 2000 counts. But, the onset looks good, and that, because of the need to model time zero, is the important part.
Below is a record of the drift in the downhole corrected 206/238 signal throughout the day, represented by ~70 measurements of the FC-1 zircon. I plot the signal ratios as ages because I think it's easier to interpret in that form. I plot the downhole corrected ages instead of the raw ages because the raw age just averages the signal through time and the downhole plots the ratio at time zero. This should improve consistency. The accepted age of FC-1 is ~1099 Ma.
Below is time evolved plots of two other standards, 91500 and Plesovice:
The shared long term pattern in these records makes me comfortable calling this variability instrumental drift. Iolite corrects this using a spline fit of the primary standard. You can control how intensely you want to fit the data with spline. I err towards the least complex fit, with the smoothing factor set at 10 out of 10. This assumes that most of the interstandard variability has to do with the sample and the laser rather than the mass spectrometer. The spline and the downhole corrected standards are pictured below.
I will walk through and comment on each sample with regards to the quality of the analyses, problems, and potential sources of error for sample. This section contains a lot of graphics: conchordia plots, 6-8 (206Pb/238U) age plots, 7-5 (207Pb/235U) age plots, and 7-6 (207Pb/206Pb) age plots (where appropriate). All horizontal bars on the age plots are 2 standard error of the mean for the population (2SE). Scroll to the end for tabulated results, where you can see absolute and percent 2SE as well as averaged instrumental (internal/analytical) and population propogated uncertainties (both absolute and percent).
The fact that the age here is reproduced well does not have much meaning, considering this is our primary standard. When fitting the spline through this data, I tried to make the resulting ages have a MSWD of ~1, meaning that the distribution of error in the means of the sample ages has about the same variance as would be predicted by the analytical uncertainty. Maybe this is not the way to do this, as there is instrumental drift mixed in here as a reason for the sample to sample variation. Feedback welcomed.
It looks like the ages get a bit more variable at the end. I think I was lasering kind of a crummy crystal.
The concordia age is too young and the correct age is outside of our reported uncertainty. However, the accepted age is only 1.5% higher than our mean age, and only 0.5% higher than the highest age allowed by our uncertainty. That could be worse!
If you look closesly at the 6-8 panel, it appears that it is our 206Pb/238U measurements/ages that are too low/young. If you don't want to squint, you can look at the tabulated results further below.
The 7-6 ages calculated for this zircon look awful (around 1130 Ma), so I didn't include them. Does that mean that FC-1, with a similar age, is a bad standard for Pb-Pb ages?
Once again here, our mean age for concordia age is too young. In this case it's about 5% (2.5 Ma). The max age permitted by our uncertainty is 55 Ma, which is 1% away from the correct age. Our precision overall with the concordia age is low, at 5%. However, the zircon has relatively low U (only 100k counts). That compounded with the fact that the zircon is fairly young (55 Ma) leads to extremely low Pb concentrations (~500 counts) and doubly difficult measurement. 207 counts are not distinguishable from background within error.
I ordered some crystals of Mud Tank because I the grains are extremely large, (cm scale) and I thought it might be a good target for some tests. The zircon comes from the Mud Tank carbonatite in Australia (~731 Ma). It's not useful as a primary standard for two reasons 1) there's a fair degree of crystal-crystal age variability. 2) low uranium content increases the challenge of dating it. Check out Gain et al. 2019 in Geostandards for more info.
I recieved three grains. Two were pinkish, and this one was more black-ish. I chose the blackish one, thinking that might be radiation damage and it might have the highest U concentration. The concentrations of U in the Mud Tank grain I loaded is extremely low, yielding only 25k counts. The Pb counts are accordingly low, at 2500. Consequently, the ages we derived aren't great in that they are neither accurate nor precise.
The one smaller ellipse is when I cranked up the spot size to 100 um. In retrospect, I should have done that (or even bigger!) for all my Mud Tank analyses. We have no shortage of analyte. But perplexingly, the age is still young for the Mud Tank carbonatite.
Admittedly, the I bought the crystals on an internet mineral retailer, not some kind of geology standards distributer. They were only \$40. Maybe this is a problem for the age. But the fact that this is indeed a zircon, and they wouldn't send me a huge zircon from the wrong pegmatite makes this unlikely.
My other thought was that perhaps, because counts are so low, there could be some problem with the beam seconds threshold. If the threshold is too high, your time zero will come too early. But an early time zero will make your age older, not younger. So that doesn't explain our problem. So this is still a mystery to me.
Every grain of OG-1 I analyzed is plotted below.
There seems to be at a grain of OG-1 suffering some Pb loss, or simply a bad analysis. Acknowledging this problematic measurement, I drop it for the remaining statistics.
Our concordia age is .8% lower than the accepted age, and the max age allowed by uncertainty is .6% lower. Our internal precision is quite good, which should be expected in an older sample.
For this sample, it looks like most of the 6-8 ages are accurate within uncertainty (correct age 3465 Ma), and it's the 7-5 ages that are making it a bit young.
The 7-6 ages are excellent, which should be perhaps expected with an old grain.
Every grain of Oracle I analyzed is plotted here.
The position of the two grains in the lower left definitely make me suspicious of Pb loss. There is also a bad analysis near the bottom of the cluster that has quite a large error envelope. Oracle isn't a widely accepted standard yet, so maybe this shouldn't surprise me. Acknowledging these bad measurements, I drop those 3 analyses for the remaining presentation.
The concordia age is to young by 1.7% (25 Ma), the max age allowed by our uncertainty is 1% too young.
The results here are similar to 91500 in that the 6-8 ages are too young (though a good number of samples overlap within uncertainty), but the 7-5 ages and 7-6 ages are older and better.
The concordia age for Plesovice is within 1% of the accepted age, but young. The correct age is < 1 Ma from our reported uncerainty.
The results here are similar to those for 91500 and Oracle. The 6-8 age is young, and the 7-5 age is pretty good. The good news is that all of these are overlapping within uncertainty.
These are all of the samples of Tan Br I analysed. The mean age is the little red circle.
As you can see, we found a beautiful discord in the sample, marking a Pb loss event ~600 Ma. While this is cool, we could have avoided this if we had good imaging, and it shows how many of the zircons we're measuring for unknowns could have concordant zones.
Discounting the obviously discordant samples, we measured a concordia age within 1% of the accepted age for the sample. Our uncertainty could put us within 0.5%.
The 7-6 ages are exceptional: accurate and precise. The situation with the 6-8 and 7-5 ages is similar to the other zircon standards, the 6-8 is yielding younger ages than the 7-5 ages. Though this time, both are too young. The propogated uncertainty mostly overlap with the accepted age for the 6-8 age population because it's a bit larger than the uncertainty for the 7-5 age population.
91500, 9435, OG1, Oracle, Plesovice, Tan Br, and Temora2 are too young. Population standard error checks out at <2%. So does internal precision, with the exception of low U Mud Tank and young 9435. Temora 2 is close at 2.32% Propogated population uncertainty is more of a problem. The extra error bars added on with population statistics tack on an additional %1.5-2 to all our measurements.
Of the 7-5 ages, 9435 is much to old. OG-1's 7-5 age is slightly older than its 6-8 age. Tan Br's and Oracle are still 1% too young. The other samples however, 91500, Plesovice, and Temora2, are quite accurate, all < 1%.
So 91500's 7-6 age is quite old, and given it is about the same age as FC-1, it might not make a great a standard for that purpose. One possiblity for its poor performance though is lower U and Pb abundance than FC-1. 91500 concentrations are very consistent, but low, with 206Pb counts around 10000. 207Pb only had 1000 counts. FC-1 had highly variable 206Pb counts, clustering bimodally between 25,000-125,000, but were uniformly higher than 91500. This could produce more precise 7-6 ages.
I think the Pb-Pb ages for Oracle, OG-1, and Tan Br are good. They are typically accurate within 0.3%
Iolite has two ways of reporting uncertainty. It can use the stability of the signal to calculate an "Internal" uncertainty. It can also use statistics of a population to tack on extra uncertainty if the internal uncertainty doesn't match explain the total variability in the population.
Below is the depiction of the analytical precision from the last time I did a test on the standards. The bar plots are stacked, and each standard is a different color. Because we tested both the 20 and the 40 um spots in this experiment, we don't have a ton of data to go off of. Also, for this experiment, I only reported the internal uncertainty.
Uncertainty for 6/8 ages range from 1.5-7%. For 7/5 ages, uncertainty ranged from 1-6%. Errors for 7/6 ages are mostly low, 0-2.5%, except for 91500, which is only ~1 Ga.
The results of the July 2020 experiment are below.
In this panel, I also report propogated uncertainty in the right column.
For 6/8 ages, 9435 (aged 55 Ma) and Mud Tank (low U) have high uncertainties. Apart from these however, most zircon has internal precision of <1-3%. As mentioned when we were looking at the table, the propogated uncertainty tacks on 1.5-2% uncertainty at a minimum. This means that our analytical uncertainties were underestimating our real uncertainty.
You actually can't see Mud Tank and 9435 because the axes are limited between 0-10, and their uncertainties fall out that range. Temora and Plesovice also have internal uncertainties that don't look too good. This is I guess because 207 is 100x less abundant all of these samples. The distribution of propogated uncertainty looks approximately the same for 7-5. That means that our instrumental uncertainty is a pretty good assessment of our real uncertainty.
Some of the Pb-Pb ages, even for our standards, doesn't look good at all. The precision for our very old samples, OG1 and TanBr, is excellent though.
I think that the error is systematic enough that it could be indicating a problem with the Pb-U ratios on file for FC-1. Its a file that I made myself, so this is possible, but I have checked all of this extensively. I used both Pb-U ratios from Mattinson, 2010, Chemical Geology, which are slightly younger, and theoretical ratios I calculated myself from an age of 1099 Ma. The ratios aren't identical but they are close, and using both yields ages that are too young. If I use 91500, a very common reference standard, the age of FC-1 comes out to ~1108. I could generate figures for all the other zircons, using that zircon as the standard. Using 91500 seems to fix some things but create it's own problems.
The fact that our propogated uncertainty in 206/238 is larger than our analytical uncertainty might also point to some problem with 206/238 we're missing that skews our ages young.
But I've really bashed my head against this and can't think of anything different to try besides just changing the 206/238 isotope ratio in the standards file to make all the other ages come out right. But I hope there's a better answer. I also am not sure how to bring error down moreso than it already is. I tried to apply everything I learned and might need time to think of more things to try. Longer dwell times might help. I hope also, that feedback from you all will help open up some perspective that I so far have been missing.
I guess I shouldn't despair too much. A lot of these ages are really close!